System Identification using Support Vector Regression

With the advantage of fast computation in recent times, numerical model simulation has gained a lot of research attention. Numerical models are now widely being used for the preliminary design and analysis of ship models without actually having to perform field trials, which are both expensive and time-consuming. This project work aims to construct an Abkowitz ship hydrodynamics numerical model by deriving the ship hydrodynamic coefficients. Regression techniques are used to identify the hydrodynamic coefficients from the training data generated using simulations. This study specifically implements support vector regression to regress the coefficients from the training data. The identified coefficients are then validated by comparing the simulated trajectories using both the identified coefficients and experimentally identified coefficients.

The International Maritime Organization specifies a set of standards to be followed in the initial stages of a ship design. Explicit importance regarding ship maneuverability is mentioned in the rules and hence its study becomes important in the design process. Different methods including empirical formula-based methods, free-running model tests, and computer-based simulations are known and used. From the above methods, computer-based simulations are popular due to their easy and feasible applications. To perform computer-based simulations accurate mathematical modeling is a necessity. Hence, identifying an underlying mathematical model behind the physical process becomes an important work.

Some of the popular mathematical models include the Abkowitz and the MMG model which try to capture the underlying dynamics of the physical system using hydrodynamic coefficients. These hydrodynamic coefficients try to represent the hull-water interaction during a maneuver. Hence, identifying these coefficients becomes much important for a reliable computer simulation study. Traditionally, physical experiments using facilities like the towing tank, and planar motion mechanism (PMM) facility to get the coefficients have been explored. While these methods are quite reliable, they are quite expensive and sometimes infeasible. Due to these reasons, system identification using free-running models has been explored.